More correction of GPU classification as integrated/discrete type for DX12 [UMA flag]#254
Conversation
kolkov
left a comment
kolkov
left a comment
There was a problem hiding this comment.
Great work! This is exactly the UMA approach we suggested in PR #252. The implementation matches Rust wgpu — verified against wgpu-hal/src/dx12/adapter.rs:108-117 (UMA query) and lines 153-161 (device type classification).
Verified
D3D12_FEATURE_DATA_ARCHITECTUREstruct layout:{NodeIndex uint32, TileBasedRenderer int32, UMA int32, CacheCoherentUMA int32}= 16 bytes, ABI-correct (Win32 BOOL = 4 bytes =int32)CheckFeatureSupport(D3D12_FEATURE_ARCHITECTURE, ...)— correct constant, correct sizeof- Error guard:
if err == nil && arch.UMA != 0— on failure, defaults to Discrete (correct fallback) - Both
AdapterandAdapterLegacyupdated — no desync deviceType()logic: SOFTWARE → CPU, UMA → Integrated, else → Discrete — exact Rust wgpu match
Requested changes (two small additions for Rust wgpu parity)
1. Store CacheCoherentUMA (required for full Rust wgpu parity)
The struct already has the field — free information from the same CheckFeatureSupport call. Rust wgpu stores it as MemoryArchitecture::Unified { cache_coherent } and uses it for memory pool selection (device.rs:91-93 — D3D12_MEMORY_POOL_L0 for UMA vs L1 for non-UMA). Please add:
// In AdapterCapabilities:
IsCacheCoherentUMA bool
// In queryArchitecture (Adapter):
a.capabilities.IsCacheCoherentUMA = arch.CacheCoherentUMA != 0
// Same for AdapterLegacy inline block2. Add diagnostic logging (required for consistency)
The existing queryD3D12Options logs queried values via hal.Logger().Info(...). Architecture query should follow the same pattern for debuggability:
hal.Logger().Info("dx12: adapter architecture",
"uma", a.capabilities.IsUMA,
"cacheCoherentUMA", arch.CacheCoherentUMA != 0,
"tileBasedRenderer", arch.TileBasedRenderer != 0,
)Both changes are small — a few lines each. Once added, this is ready to merge. Thank you for implementing the UMA approach from our PR #252 feedback!
Store IsCacheCoherentUMA from D3D12_FEATURE_DATA_ARCHITECTURE for future memory pool selection (D3D12_MEMORY_POOL_L0 vs L1). Add hal.Logger().Info() for architecture query — matches existing queryD3D12Options logging pattern. Both Adapter and AdapterLegacy.
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
kolkov
left a comment
kolkov
left a comment
There was a problem hiding this comment.
Changes implemented: CacheCoherentUMA + diagnostic logging. LGTM.
2 participants
Continue #252
sorry i cant renew old PR :)
I tested all OK. my video now correctly show as Integrated. On RDP also detect as right GPU type.